1 /*
   2  * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include <jni.h>
  26 #include <unistd.h>
  27 #include <fcntl.h>
  28 #include <string.h>
  29 #include <stdlib.h>
  30 #include <stddef.h>
  31 #include <elf.h>
  32 #include <link.h>
  33 #include "libproc_impl.h"
  34 #include "salibelf.h"
  35 
  36 // This file has the libproc implementation to read core files.
  37 // For live processes, refer to ps_proc.c. Portions of this is adapted
  38 // /modelled after Solaris libproc.so (in particular Pcore.c)
  39 
  40 //----------------------------------------------------------------------
  41 // ps_prochandle cleanup helper functions
  42 
  43 // close all file descriptors
  44 static void close_files(struct ps_prochandle* ph) {
  45   lib_info* lib = NULL;
  46 
  47   // close core file descriptor
  48   if (ph->core->core_fd >= 0)
  49     close(ph->core->core_fd);
  50 
  51   // close exec file descriptor
  52   if (ph->core->exec_fd >= 0)
  53     close(ph->core->exec_fd);
  54 
  55   // close interp file descriptor
  56   if (ph->core->interp_fd >= 0)
  57     close(ph->core->interp_fd);
  58 
  59   // close class share archive file
  60   if (ph->core->classes_jsa_fd >= 0)
  61     close(ph->core->classes_jsa_fd);
  62 
  63   // close all library file descriptors
  64   lib = ph->libs;
  65   while (lib) {
  66     int fd = lib->fd;
  67     if (fd >= 0 && fd != ph->core->exec_fd) {
  68       close(fd);
  69     }
  70     lib = lib->next;
  71   }
  72 }
  73 
  74 // clean all map_info stuff
  75 static void destroy_map_info(struct ps_prochandle* ph) {
  76   map_info* map = ph->core->maps;
  77   while (map) {
  78     map_info* next = map->next;
  79     free(map);
  80     map = next;
  81   }
  82 
  83   if (ph->core->map_array) {
  84     free(ph->core->map_array);
  85   }
  86 
  87   // Part of the class sharing workaround
  88   map = ph->core->class_share_maps;
  89   while (map) {
  90     map_info* next = map->next;
  91     free(map);
  92     map = next;
  93   }
  94 }
  95 
  96 // ps_prochandle operations
  97 static void core_release(struct ps_prochandle* ph) {
  98   if (ph->core) {
  99     close_files(ph);
 100     destroy_map_info(ph);
 101     free(ph->core);
 102   }
 103 }
 104 
 105 static map_info* allocate_init_map(int fd, off_t offset, uintptr_t vaddr, size_t memsz) {
 106   map_info* map;
 107   if ( (map = (map_info*) calloc(1, sizeof(map_info))) == NULL) {
 108     print_debug("can't allocate memory for map_info\n");
 109     return NULL;
 110   }
 111 
 112   // initialize map
 113   map->fd     = fd;
 114   map->offset = offset;
 115   map->vaddr  = vaddr;
 116   map->memsz  = memsz;
 117   return map;
 118 }
 119 
 120 // add map info with given fd, offset, vaddr and memsz
 121 static map_info* add_map_info(struct ps_prochandle* ph, int fd, off_t offset,
 122                              uintptr_t vaddr, size_t memsz) {
 123   map_info* map;
 124   if ((map = allocate_init_map(fd, offset, vaddr, memsz)) == NULL) {
 125     return NULL;
 126   }
 127 
 128   // add this to map list
 129   map->next  = ph->core->maps;
 130   ph->core->maps   = map;
 131   ph->core->num_maps++;
 132 
 133   return map;
 134 }
 135 
 136 // Part of the class sharing workaround
 137 static map_info* add_class_share_map_info(struct ps_prochandle* ph, off_t offset,
 138                              uintptr_t vaddr, size_t memsz) {
 139   map_info* map;
 140   if ((map = allocate_init_map(ph->core->classes_jsa_fd,
 141                                offset, vaddr, memsz)) == NULL) {
 142     return NULL;
 143   }
 144 
 145   map->next = ph->core->class_share_maps;
 146   ph->core->class_share_maps = map;
 147   return map;
 148 }
 149 
 150 // Return the map_info for the given virtual address.  We keep a sorted
 151 // array of pointers in ph->map_array, so we can binary search.
 152 static map_info* core_lookup(struct ps_prochandle *ph, uintptr_t addr) {
 153   int mid, lo = 0, hi = ph->core->num_maps - 1;
 154   map_info *mp;
 155 
 156   while (hi - lo > 1) {
 157     mid = (lo + hi) / 2;
 158     if (addr >= ph->core->map_array[mid]->vaddr) {
 159       lo = mid;
 160     } else {
 161       hi = mid;
 162     }
 163   }
 164 
 165   if (addr < ph->core->map_array[hi]->vaddr) {
 166     mp = ph->core->map_array[lo];
 167   } else {
 168     mp = ph->core->map_array[hi];
 169   }
 170 
 171   if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) {
 172     return (mp);
 173   }
 174 
 175 
 176   // Part of the class sharing workaround
 177   // Unfortunately, we have no way of detecting -Xshare state.
 178   // Check out the share maps atlast, if we don't find anywhere.
 179   // This is done this way so to avoid reading share pages
 180   // ahead of other normal maps. For eg. with -Xshare:off we don't
 181   // want to prefer class sharing data to data from core.
 182   mp = ph->core->class_share_maps;
 183   if (mp) {
 184     print_debug("can't locate map_info at 0x%lx, trying class share maps\n", addr);
 185   }
 186   while (mp) {
 187     if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) {
 188       print_debug("located map_info at 0x%lx from class share maps\n", addr);
 189       return (mp);
 190     }
 191     mp = mp->next;
 192   }
 193 
 194   print_debug("can't locate map_info at 0x%lx\n", addr);
 195   return (NULL);
 196 }
 197 
 198 //---------------------------------------------------------------
 199 // Part of the class sharing workaround:
 200 //
 201 // With class sharing, pages are mapped from classes.jsa file.
 202 // The read-only class sharing pages are mapped as MAP_SHARED,
 203 // PROT_READ pages. These pages are not dumped into core dump.
 204 // With this workaround, these pages are read from classes.jsa.
 205 
 206 // FIXME: !HACK ALERT!
 207 // The format of sharing achive file header is needed to read shared heap
 208 // file mappings. For now, I am hard coding portion of FileMapHeader here.
 209 // Refer to filemap.hpp.
 210 
 211 // FileMapHeader describes the shared space data in the file to be
 212 // mapped.  This structure gets written to a file.  It is not a class,
 213 // so that the compilers don't add any compiler-private data to it.
 214 
 215 #define NUM_SHARED_MAPS 4
 216 
 217 // Refer to FileMapInfo::_current_version in filemap.hpp
 218 #define CURRENT_ARCHIVE_VERSION 1
 219 
 220 struct FileMapHeader {
 221   int   _magic;              // identify file type.
 222   int   _version;            // (from enum, above.)
 223   size_t _alignment;         // how shared archive should be aligned
 224 
 225   struct space_info {
 226     int    _file_offset;     // sizeof(this) rounded to vm page size
 227     char*  _base;            // copy-on-write base address
 228     size_t _capacity;        // for validity checking
 229     size_t _used;            // for setting space top on read
 230 
 231     // 4991491 NOTICE These are C++ bool's in filemap.hpp and must match up with
 232     // the C type matching the C++ bool type on any given platform.
 233     // We assume the corresponding C type is char but licensees
 234     // may need to adjust the type of these fields.
 235     char   _read_only;       // read only space?
 236     char   _allow_exec;      // executable code in space?
 237 
 238   } _space[NUM_SHARED_MAPS];
 239 
 240   // Ignore the rest of the FileMapHeader. We don't need those fields here.
 241 };
 242 
 243 static bool read_jboolean(struct ps_prochandle* ph, uintptr_t addr, jboolean* pvalue) {
 244   jboolean i;
 245   if (ps_pdread(ph, (psaddr_t) addr, &i, sizeof(i)) == PS_OK) {
 246     *pvalue = i;
 247     return true;
 248   } else {
 249     return false;
 250   }
 251 }
 252 
 253 static bool read_pointer(struct ps_prochandle* ph, uintptr_t addr, uintptr_t* pvalue) {
 254   uintptr_t uip;
 255   if (ps_pdread(ph, (psaddr_t) addr, (char *)&uip, sizeof(uip)) == PS_OK) {
 256     *pvalue = uip;
 257     return true;
 258   } else {
 259     return false;
 260   }
 261 }
 262 
 263 // used to read strings from debuggee
 264 static bool read_string(struct ps_prochandle* ph, uintptr_t addr, char* buf, size_t size) {
 265   size_t i = 0;
 266   char  c = ' ';
 267 
 268   while (c != '\0') {
 269     if (ps_pdread(ph, (psaddr_t) addr, &c, sizeof(char)) != PS_OK) {
 270       return false;
 271     }
 272     if (i < size - 1) {
 273       buf[i] = c;
 274     } else {
 275       // smaller buffer
 276       return false;
 277     }
 278     i++; addr++;
 279   }
 280 
 281   buf[i] = '\0';
 282   return true;
 283 }
 284 
 285 #define USE_SHARED_SPACES_SYM "UseSharedSpaces"
 286 // mangled name of Arguments::SharedArchivePath
 287 #define SHARED_ARCHIVE_PATH_SYM "_ZN9Arguments17SharedArchivePathE"
 288 #define LIBJVM_NAME "/libjvm.so"
 289 
 290 static bool init_classsharing_workaround(struct ps_prochandle* ph) {
 291   lib_info* lib = ph->libs;
 292   while (lib != NULL) {
 293     // we are iterating over shared objects from the core dump. look for
 294     // libjvm.so.
 295     const char *jvm_name = 0;
 296     if ((jvm_name = strstr(lib->name, LIBJVM_NAME)) != 0) {
 297       char classes_jsa[PATH_MAX];
 298       struct FileMapHeader header;
 299       int fd = -1;
 300       int m = 0;
 301       size_t n = 0;
 302       uintptr_t base = 0, useSharedSpacesAddr = 0;
 303       uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0;
 304       jboolean useSharedSpaces = 0;
 305       map_info* mi = 0;
 306 
 307       memset(classes_jsa, 0, sizeof(classes_jsa));
 308       jvm_name = lib->name;
 309       useSharedSpacesAddr = lookup_symbol(ph, jvm_name, USE_SHARED_SPACES_SYM);
 310       if (useSharedSpacesAddr == 0) {
 311         print_debug("can't lookup 'UseSharedSpaces' flag\n");
 312         return false;
 313       }
 314 
 315       // Hotspot vm types are not exported to build this library. So
 316       // using equivalent type jboolean to read the value of
 317       // UseSharedSpaces which is same as hotspot type "bool".
 318       if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) {
 319         print_debug("can't read the value of 'UseSharedSpaces' flag\n");
 320         return false;
 321       }
 322 
 323       if ((int)useSharedSpaces == 0) {
 324         print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n");
 325         return true;
 326       }
 327 
 328       sharedArchivePathAddrAddr = lookup_symbol(ph, jvm_name, SHARED_ARCHIVE_PATH_SYM);
 329       if (sharedArchivePathAddrAddr == 0) {
 330         print_debug("can't lookup shared archive path symbol\n");
 331         return false;
 332       }
 333 
 334       if (read_pointer(ph, sharedArchivePathAddrAddr, &sharedArchivePathAddr) != true) {
 335         print_debug("can't read shared archive path pointer\n");
 336         return false;
 337       }
 338 
 339       if (read_string(ph, sharedArchivePathAddr, classes_jsa, sizeof(classes_jsa)) != true) {
 340         print_debug("can't read shared archive path value\n");
 341         return false;
 342       }
 343 
 344       print_debug("looking for %s\n", classes_jsa);
 345       // open the class sharing archive file
 346       fd = pathmap_open(classes_jsa);
 347       if (fd < 0) {
 348         print_debug("can't open %s!\n", classes_jsa);
 349         ph->core->classes_jsa_fd = -1;
 350         return false;
 351       } else {
 352         print_debug("opened %s\n", classes_jsa);
 353       }
 354 
 355       // read FileMapHeader from the file
 356       memset(&header, 0, sizeof(struct FileMapHeader));
 357       if ((n = read(fd, &header, sizeof(struct FileMapHeader)))
 358            != sizeof(struct FileMapHeader)) {
 359         print_debug("can't read shared archive file map header from %s\n", classes_jsa);
 360         close(fd);
 361         return false;
 362       }
 363 
 364       // check file magic
 365       if (header._magic != 0xf00baba2) {
 366         print_debug("%s has bad shared archive file magic number 0x%x, expecing 0xf00baba2\n",
 367                      classes_jsa, header._magic);
 368         close(fd);
 369         return false;
 370       }
 371 
 372       // check version
 373       if (header._version != CURRENT_ARCHIVE_VERSION) {
 374         print_debug("%s has wrong shared archive file version %d, expecting %d\n",
 375                      classes_jsa, header._version, CURRENT_ARCHIVE_VERSION);
 376         close(fd);
 377         return false;
 378       }
 379 
 380       ph->core->classes_jsa_fd = fd;
 381       // add read-only maps from classes.jsa to the list of maps
 382       for (m = 0; m < NUM_SHARED_MAPS; m++) {
 383         if (header._space[m]._read_only) {
 384           base = (uintptr_t) header._space[m]._base;
 385           // no need to worry about the fractional pages at-the-end.
 386           // possible fractional pages are handled by core_read_data.
 387           add_class_share_map_info(ph, (off_t) header._space[m]._file_offset,
 388                                    base, (size_t) header._space[m]._used);
 389           print_debug("added a share archive map at 0x%lx\n", base);
 390         }
 391       }
 392       return true;
 393    }
 394    lib = lib->next;
 395   }
 396   return true;
 397 }
 398 
 399 
 400 //---------------------------------------------------------------------------
 401 // functions to handle map_info
 402 
 403 // Order mappings based on virtual address.  We use this function as the
 404 // callback for sorting the array of map_info pointers.
 405 static int core_cmp_mapping(const void *lhsp, const void *rhsp)
 406 {
 407   const map_info *lhs = *((const map_info **)lhsp);
 408   const map_info *rhs = *((const map_info **)rhsp);
 409 
 410   if (lhs->vaddr == rhs->vaddr) {
 411     return (0);
 412   }
 413 
 414   return (lhs->vaddr < rhs->vaddr ? -1 : 1);
 415 }
 416 
 417 // we sort map_info by starting virtual address so that we can do
 418 // binary search to read from an address.
 419 static bool sort_map_array(struct ps_prochandle* ph) {
 420   size_t num_maps = ph->core->num_maps;
 421   map_info* map = ph->core->maps;
 422   int i = 0;
 423 
 424   // allocate map_array
 425   map_info** array;
 426   if ( (array = (map_info**) malloc(sizeof(map_info*) * num_maps)) == NULL) {
 427     print_debug("can't allocate memory for map array\n");
 428     return false;
 429   }
 430 
 431   // add maps to array
 432   while (map) {
 433     array[i] = map;
 434     i++;
 435     map = map->next;
 436   }
 437 
 438   // sort is called twice. If this is second time, clear map array
 439   if (ph->core->map_array) {
 440     free(ph->core->map_array);
 441   }
 442 
 443   ph->core->map_array = array;
 444   // sort the map_info array by base virtual address.
 445   qsort(ph->core->map_array, ph->core->num_maps, sizeof (map_info*),
 446         core_cmp_mapping);
 447 
 448   // print map
 449   if (is_debug()) {
 450     int j = 0;
 451     print_debug("---- sorted virtual address map ----\n");
 452     for (j = 0; j < ph->core->num_maps; j++) {
 453       print_debug("base = 0x%lx\tsize = %zu\n", ph->core->map_array[j]->vaddr,
 454                   ph->core->map_array[j]->memsz);
 455     }
 456   }
 457 
 458   return true;
 459 }
 460 
 461 #ifndef MIN
 462 #define MIN(x, y) (((x) < (y))? (x): (y))
 463 #endif
 464 
 465 static bool core_read_data(struct ps_prochandle* ph, uintptr_t addr, char *buf, size_t size) {
 466    ssize_t resid = size;
 467    int page_size=sysconf(_SC_PAGE_SIZE);
 468    while (resid != 0) {
 469       map_info *mp = core_lookup(ph, addr);
 470       uintptr_t mapoff;
 471       ssize_t len, rem;
 472       off_t off;
 473       int fd;
 474 
 475       if (mp == NULL) {
 476          break;  /* No mapping for this address */
 477       }
 478 
 479       fd = mp->fd;
 480       mapoff = addr - mp->vaddr;
 481       len = MIN(resid, mp->memsz - mapoff);
 482       off = mp->offset + mapoff;
 483 
 484       if ((len = pread(fd, buf, len, off)) <= 0) {
 485          break;
 486       }
 487 
 488       resid -= len;
 489       addr += len;
 490       buf = (char *)buf + len;
 491 
 492       // mappings always start at page boundary. But, may end in fractional
 493       // page. fill zeros for possible fractional page at the end of a mapping.
 494       rem = mp->memsz % page_size;
 495       if (rem > 0) {
 496          rem = page_size - rem;
 497          len = MIN(resid, rem);
 498          resid -= len;
 499          addr += len;
 500          // we are not assuming 'buf' to be zero initialized.
 501          memset(buf, 0, len);
 502          buf += len;
 503       }
 504    }
 505 
 506    if (resid) {
 507       print_debug("core read failed for %d byte(s) @ 0x%lx (%d more bytes)\n",
 508               size, addr, resid);
 509       return false;
 510    } else {
 511       return true;
 512    }
 513 }
 514 
 515 // null implementation for write
 516 static bool core_write_data(struct ps_prochandle* ph,
 517                              uintptr_t addr, const char *buf , size_t size) {
 518    return false;
 519 }
 520 
 521 static bool core_get_lwp_regs(struct ps_prochandle* ph, lwpid_t lwp_id,
 522                           struct user_regs_struct* regs) {
 523    // for core we have cached the lwp regs from NOTE section
 524    thread_info* thr = ph->threads;
 525    while (thr) {
 526      if (thr->lwp_id == lwp_id) {
 527        memcpy(regs, &thr->regs, sizeof(struct user_regs_struct));
 528        return true;
 529      }
 530      thr = thr->next;
 531    }
 532    return false;
 533 }
 534 
 535 static ps_prochandle_ops core_ops = {
 536    .release=  core_release,
 537    .p_pread=  core_read_data,
 538    .p_pwrite= core_write_data,
 539    .get_lwp_regs= core_get_lwp_regs
 540 };
 541 
 542 // read regs and create thread from NT_PRSTATUS entries from core file
 543 static bool core_handle_prstatus(struct ps_prochandle* ph, const char* buf, size_t nbytes) {
 544    // we have to read prstatus_t from buf
 545    // assert(nbytes == sizeof(prstaus_t), "size mismatch on prstatus_t");
 546    prstatus_t* prstat = (prstatus_t*) buf;
 547    thread_info* newthr;
 548    print_debug("got integer regset for lwp %d\n", prstat->pr_pid);
 549    // we set pthread_t to -1 for core dump
 550    if((newthr = add_thread_info(ph, (pthread_t) -1,  prstat->pr_pid)) == NULL)
 551       return false;
 552 
 553    // copy regs
 554    memcpy(&newthr->regs, prstat->pr_reg, sizeof(struct user_regs_struct));
 555 
 556    if (is_debug()) {
 557       print_debug("integer regset\n");
 558 #ifdef i386
 559       // print the regset
 560       print_debug("\teax = 0x%x\n", newthr->regs.eax);
 561       print_debug("\tebx = 0x%x\n", newthr->regs.ebx);
 562       print_debug("\tecx = 0x%x\n", newthr->regs.ecx);
 563       print_debug("\tedx = 0x%x\n", newthr->regs.edx);
 564       print_debug("\tesp = 0x%x\n", newthr->regs.esp);
 565       print_debug("\tebp = 0x%x\n", newthr->regs.ebp);
 566       print_debug("\tesi = 0x%x\n", newthr->regs.esi);
 567       print_debug("\tedi = 0x%x\n", newthr->regs.edi);
 568       print_debug("\teip = 0x%x\n", newthr->regs.eip);
 569 #endif
 570 
 571 #if defined(amd64) || defined(x86_64)
 572       // print the regset
 573       print_debug("\tr15 = 0x%lx\n", newthr->regs.r15);
 574       print_debug("\tr14 = 0x%lx\n", newthr->regs.r14);
 575       print_debug("\tr13 = 0x%lx\n", newthr->regs.r13);
 576       print_debug("\tr12 = 0x%lx\n", newthr->regs.r12);
 577       print_debug("\trbp = 0x%lx\n", newthr->regs.rbp);
 578       print_debug("\trbx = 0x%lx\n", newthr->regs.rbx);
 579       print_debug("\tr11 = 0x%lx\n", newthr->regs.r11);
 580       print_debug("\tr10 = 0x%lx\n", newthr->regs.r10);
 581       print_debug("\tr9 = 0x%lx\n", newthr->regs.r9);
 582       print_debug("\tr8 = 0x%lx\n", newthr->regs.r8);
 583       print_debug("\trax = 0x%lx\n", newthr->regs.rax);
 584       print_debug("\trcx = 0x%lx\n", newthr->regs.rcx);
 585       print_debug("\trdx = 0x%lx\n", newthr->regs.rdx);
 586       print_debug("\trsi = 0x%lx\n", newthr->regs.rsi);
 587       print_debug("\trdi = 0x%lx\n", newthr->regs.rdi);
 588       print_debug("\torig_rax = 0x%lx\n", newthr->regs.orig_rax);
 589       print_debug("\trip = 0x%lx\n", newthr->regs.rip);
 590       print_debug("\tcs = 0x%lx\n", newthr->regs.cs);
 591       print_debug("\teflags = 0x%lx\n", newthr->regs.eflags);
 592       print_debug("\trsp = 0x%lx\n", newthr->regs.rsp);
 593       print_debug("\tss = 0x%lx\n", newthr->regs.ss);
 594       print_debug("\tfs_base = 0x%lx\n", newthr->regs.fs_base);
 595       print_debug("\tgs_base = 0x%lx\n", newthr->regs.gs_base);
 596       print_debug("\tds = 0x%lx\n", newthr->regs.ds);
 597       print_debug("\tes = 0x%lx\n", newthr->regs.es);
 598       print_debug("\tfs = 0x%lx\n", newthr->regs.fs);
 599       print_debug("\tgs = 0x%lx\n", newthr->regs.gs);
 600 #endif
 601    }
 602 
 603    return true;
 604 }
 605 
 606 #define ROUNDUP(x, y)  ((((x)+((y)-1))/(y))*(y))
 607 
 608 // read NT_PRSTATUS entries from core NOTE segment
 609 static bool core_handle_note(struct ps_prochandle* ph, ELF_PHDR* note_phdr) {
 610    char* buf = NULL;
 611    char* p = NULL;
 612    size_t size = note_phdr->p_filesz;
 613 
 614    // we are interested in just prstatus entries. we will ignore the rest.
 615    // Advance the seek pointer to the start of the PT_NOTE data
 616    if (lseek(ph->core->core_fd, note_phdr->p_offset, SEEK_SET) == (off_t)-1) {
 617       print_debug("failed to lseek to PT_NOTE data\n");
 618       return false;
 619    }
 620 
 621    // Now process the PT_NOTE structures.  Each one is preceded by
 622    // an Elf{32/64}_Nhdr structure describing its type and size.
 623    if ( (buf = (char*) malloc(size)) == NULL) {
 624       print_debug("can't allocate memory for reading core notes\n");
 625       goto err;
 626    }
 627 
 628    // read notes into buffer
 629    if (read(ph->core->core_fd, buf, size) != size) {
 630       print_debug("failed to read notes, core file must have been truncated\n");
 631       goto err;
 632    }
 633 
 634    p = buf;
 635    while (p < buf + size) {
 636       ELF_NHDR* notep = (ELF_NHDR*) p;
 637       char* descdata  = p + sizeof(ELF_NHDR) + ROUNDUP(notep->n_namesz, 4);
 638       print_debug("Note header with n_type = %d and n_descsz = %u\n",
 639                                    notep->n_type, notep->n_descsz);
 640 
 641       if (notep->n_type == NT_PRSTATUS) {
 642         if (core_handle_prstatus(ph, descdata, notep->n_descsz) != true) {
 643           return false;
 644         }
 645       } else if (notep->n_type == NT_AUXV) {
 646         // Get first segment from entry point
 647         ELF_AUXV *auxv = (ELF_AUXV *)descdata;
 648         while (auxv->a_type != AT_NULL) {
 649           if (auxv->a_type == AT_ENTRY) {
 650             // Set entry point address to address of dynamic section.
 651             // We will adjust it in read_exec_segments().
 652             ph->core->dynamic_addr = auxv->a_un.a_val;
 653             break;
 654           }
 655           auxv++;
 656         }
 657       }
 658       p = descdata + ROUNDUP(notep->n_descsz, 4);
 659    }
 660 
 661    free(buf);
 662    return true;
 663 
 664 err:
 665    if (buf) free(buf);
 666    return false;
 667 }
 668 
 669 // read all segments from core file
 670 static bool read_core_segments(struct ps_prochandle* ph, ELF_EHDR* core_ehdr) {
 671    int i = 0;
 672    ELF_PHDR* phbuf = NULL;
 673    ELF_PHDR* core_php = NULL;
 674 
 675    if ((phbuf =  read_program_header_table(ph->core->core_fd, core_ehdr)) == NULL)
 676       return false;
 677 
 678    /*
 679     * Now iterate through the program headers in the core file.
 680     * We're interested in two types of Phdrs: PT_NOTE (which
 681     * contains a set of saved /proc structures), and PT_LOAD (which
 682     * represents a memory mapping from the process's address space).
 683     *
 684     * Difference b/w Solaris PT_NOTE and Linux/BSD PT_NOTE:
 685     *
 686     *     In Solaris there are two PT_NOTE segments the first PT_NOTE (if present)
 687     *     contains /proc structs in the pre-2.6 unstructured /proc format. the last
 688     *     PT_NOTE has data in new /proc format.
 689     *
 690     *     In Solaris, there is only one pstatus (process status). pstatus contains
 691     *     integer register set among other stuff. For each LWP, we have one lwpstatus
 692     *     entry that has integer regset for that LWP.
 693     *
 694     *     Linux threads are actually 'clone'd processes. To support core analysis
 695     *     of "multithreaded" process, Linux creates more than one pstatus (called
 696     *     "prstatus") entry in PT_NOTE. Each prstatus entry has integer regset for one
 697     *     "thread". Please refer to Linux kernel src file 'fs/binfmt_elf.c', in particular
 698     *     function "elf_core_dump".
 699     */
 700 
 701     for (core_php = phbuf, i = 0; i < core_ehdr->e_phnum; i++) {
 702       switch (core_php->p_type) {
 703          case PT_NOTE:
 704             if (core_handle_note(ph, core_php) != true) {
 705               goto err;
 706             }
 707             break;
 708 
 709          case PT_LOAD: {
 710             if (core_php->p_filesz != 0) {
 711                if (add_map_info(ph, ph->core->core_fd, core_php->p_offset,
 712                   core_php->p_vaddr, core_php->p_filesz) == NULL) goto err;
 713             }
 714             break;
 715          }
 716       }
 717 
 718       core_php++;
 719    }
 720 
 721    free(phbuf);
 722    return true;
 723 err:
 724    free(phbuf);
 725    return false;
 726 }
 727 
 728 // read segments of a shared object
 729 static bool read_lib_segments(struct ps_prochandle* ph, int lib_fd, ELF_EHDR* lib_ehdr, uintptr_t lib_base) {
 730   int i = 0;
 731   ELF_PHDR* phbuf;
 732   ELF_PHDR* lib_php = NULL;
 733 
 734   int page_size = sysconf(_SC_PAGE_SIZE);
 735 
 736   if ((phbuf = read_program_header_table(lib_fd, lib_ehdr)) == NULL) {
 737     return false;
 738   }
 739 
 740   // we want to process only PT_LOAD segments that are not writable.
 741   // i.e., text segments. The read/write/exec (data) segments would
 742   // have been already added from core file segments.
 743   for (lib_php = phbuf, i = 0; i < lib_ehdr->e_phnum; i++) {
 744     if ((lib_php->p_type == PT_LOAD) && !(lib_php->p_flags & PF_W) && (lib_php->p_filesz != 0)) {
 745 
 746       uintptr_t target_vaddr = lib_php->p_vaddr + lib_base;
 747       map_info *existing_map = core_lookup(ph, target_vaddr);
 748 
 749       if (existing_map == NULL){
 750         if (add_map_info(ph, lib_fd, lib_php->p_offset,
 751                           target_vaddr, lib_php->p_memsz) == NULL) {
 752           goto err;
 753         }
 754       } else {
 755         // Coredump stores value of p_memsz elf field
 756         // rounded up to page boundary.
 757 
 758         if ((existing_map->memsz != page_size) &&
 759             (existing_map->fd != lib_fd) &&
 760             (ROUNDUP(existing_map->memsz, page_size) != ROUNDUP(lib_php->p_memsz, page_size))) {
 761 
 762           print_debug("address conflict @ 0x%lx (existing map size = %ld, size = %ld, flags = %d)\n",
 763                         target_vaddr, existing_map->memsz, lib_php->p_memsz, lib_php->p_flags);
 764           goto err;
 765         }
 766 
 767         /* replace PT_LOAD segment with library segment */
 768         print_debug("overwrote with new address mapping (memsz %ld -> %ld)\n",
 769                      existing_map->memsz, ROUNDUP(lib_php->p_memsz, page_size));
 770 
 771         existing_map->fd = lib_fd;
 772         existing_map->offset = lib_php->p_offset;
 773         existing_map->memsz = ROUNDUP(lib_php->p_memsz, page_size);
 774       }
 775     }
 776 
 777     lib_php++;
 778   }
 779 
 780   free(phbuf);
 781   return true;
 782 err:
 783   free(phbuf);
 784   return false;
 785 }
 786 
 787 // process segments from interpreter (ld.so or ld-linux.so)
 788 static bool read_interp_segments(struct ps_prochandle* ph) {
 789   ELF_EHDR interp_ehdr;
 790 
 791   if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) {
 792     print_debug("interpreter is not a valid ELF file\n");
 793     return false;
 794   }
 795 
 796   if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) {
 797     print_debug("can't read segments of interpreter\n");
 798     return false;
 799   }
 800 
 801   return true;
 802 }
 803 
 804 // process segments of a a.out
 805 static bool read_exec_segments(struct ps_prochandle* ph, ELF_EHDR* exec_ehdr) {
 806   int i = 0;
 807   ELF_PHDR* phbuf = NULL;
 808   ELF_PHDR* exec_php = NULL;
 809 
 810   if ((phbuf = read_program_header_table(ph->core->exec_fd, exec_ehdr)) == NULL) {
 811     return false;
 812   }
 813 
 814   for (exec_php = phbuf, i = 0; i < exec_ehdr->e_phnum; i++) {
 815     switch (exec_php->p_type) {
 816 
 817       // add mappings for PT_LOAD segments
 818     case PT_LOAD: {
 819       // add only non-writable segments of non-zero filesz
 820       if (!(exec_php->p_flags & PF_W) && exec_php->p_filesz != 0) {
 821         if (add_map_info(ph, ph->core->exec_fd, exec_php->p_offset, exec_php->p_vaddr, exec_php->p_filesz) == NULL) goto err;
 822       }
 823       break;
 824     }
 825 
 826     // read the interpreter and it's segments
 827     case PT_INTERP: {
 828       char interp_name[BUF_SIZE + 1];
 829 
 830       // BUF_SIZE is PATH_MAX + NAME_MAX + 1.
 831       if (exec_php->p_filesz > BUF_SIZE) {
 832         goto err;
 833       }
 834       pread(ph->core->exec_fd, interp_name, exec_php->p_filesz, exec_php->p_offset);
 835       interp_name[exec_php->p_filesz] = '\0';
 836       print_debug("ELF interpreter %s\n", interp_name);
 837       // read interpreter segments as well
 838       if ((ph->core->interp_fd = pathmap_open(interp_name)) < 0) {
 839         print_debug("can't open runtime loader\n");
 840         goto err;
 841       }
 842       break;
 843     }
 844 
 845     // from PT_DYNAMIC we want to read address of first link_map addr
 846     case PT_DYNAMIC: {
 847         if (exec_ehdr->e_type == ET_EXEC) {
 848             ph->core->dynamic_addr = exec_php->p_vaddr;
 849         } else { // ET_DYN
 850             // dynamic_addr has entry point of executable.
 851             // Thus we should substract it.
 852             ph->core->dynamic_addr += exec_php->p_vaddr - exec_ehdr->e_entry;
 853         }
 854         print_debug("address of _DYNAMIC is 0x%lx\n", ph->core->dynamic_addr);
 855         break;
 856     }
 857 
 858     } // switch
 859     exec_php++;
 860   } // for
 861 
 862   free(phbuf);
 863   return true;
 864  err:
 865   free(phbuf);
 866   return false;
 867 }
 868 
 869 
 870 #define FIRST_LINK_MAP_OFFSET offsetof(struct r_debug,  r_map)
 871 #define LD_BASE_OFFSET        offsetof(struct r_debug,  r_ldbase)
 872 #define LINK_MAP_ADDR_OFFSET  offsetof(struct link_map, l_addr)
 873 #define LINK_MAP_NAME_OFFSET  offsetof(struct link_map, l_name)
 874 #define LINK_MAP_NEXT_OFFSET  offsetof(struct link_map, l_next)
 875 
 876 // read shared library info from runtime linker's data structures.
 877 // This work is done by librtlb_db in Solaris
 878 static bool read_shared_lib_info(struct ps_prochandle* ph) {
 879   uintptr_t addr = ph->core->dynamic_addr;
 880   uintptr_t debug_base;
 881   uintptr_t first_link_map_addr;
 882   uintptr_t ld_base_addr;
 883   uintptr_t link_map_addr;
 884   uintptr_t lib_base_diff;
 885   uintptr_t lib_base;
 886   uintptr_t lib_name_addr;
 887   char lib_name[BUF_SIZE];
 888   ELF_DYN dyn;
 889   ELF_EHDR elf_ehdr;
 890   int lib_fd;
 891 
 892   // _DYNAMIC has information of the form
 893   //         [tag] [data] [tag] [data] .....
 894   // Both tag and data are pointer sized.
 895   // We look for dynamic info with DT_DEBUG. This has shared object info.
 896   // refer to struct r_debug in link.h
 897 
 898   dyn.d_tag = DT_NULL;
 899   while (dyn.d_tag != DT_DEBUG) {
 900     if (ps_pdread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) {
 901       print_debug("can't read debug info from _DYNAMIC\n");
 902       return false;
 903     }
 904     addr += sizeof(ELF_DYN);
 905   }
 906 
 907   // we have got Dyn entry with DT_DEBUG
 908   debug_base = dyn.d_un.d_ptr;
 909   // at debug_base we have struct r_debug. This has first link map in r_map field
 910   if (ps_pdread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET,
 911                  &first_link_map_addr, sizeof(uintptr_t)) != PS_OK) {
 912     print_debug("can't read first link map address\n");
 913     return false;
 914   }
 915 
 916   // read ld_base address from struct r_debug
 917   if (ps_pdread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr,
 918                  sizeof(uintptr_t)) != PS_OK) {
 919     print_debug("can't read ld base address\n");
 920     return false;
 921   }
 922   ph->core->ld_base_addr = ld_base_addr;
 923 
 924   print_debug("interpreter base address is 0x%lx\n", ld_base_addr);
 925 
 926   // now read segments from interp (i.e ld.so or ld-linux.so or ld-elf.so)
 927   if (read_interp_segments(ph) != true) {
 928       return false;
 929   }
 930 
 931   // after adding interpreter (ld.so) mappings sort again
 932   if (sort_map_array(ph) != true) {
 933     return false;
 934   }
 935 
 936    print_debug("first link map is at 0x%lx\n", first_link_map_addr);
 937 
 938    link_map_addr = first_link_map_addr;
 939    while (link_map_addr != 0) {
 940       // read library base address of the .so. Note that even though <sys/link.h> calls
 941       // link_map->l_addr as "base address",  this is * not * really base virtual
 942       // address of the shared object. This is actually the difference b/w the virtual
 943       // address mentioned in shared object and the actual virtual base where runtime
 944       // linker loaded it. We use "base diff" in read_lib_segments call below.
 945 
 946       if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET,
 947                    &lib_base_diff, sizeof(uintptr_t)) != PS_OK) {
 948          print_debug("can't read shared object base address diff\n");
 949          return false;
 950       }
 951 
 952       // read address of the name
 953       if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET,
 954                     &lib_name_addr, sizeof(uintptr_t)) != PS_OK) {
 955          print_debug("can't read address of shared object name\n");
 956          return false;
 957       }
 958 
 959       // read name of the shared object
 960       lib_name[0] = '\0';
 961       if (lib_name_addr != 0 &&
 962           read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) {
 963          print_debug("can't read shared object name\n");
 964          // don't let failure to read the name stop opening the file.  If something is really wrong
 965          // it will fail later.
 966       }
 967 
 968       if (lib_name[0] != '\0') {
 969          // ignore empty lib names
 970          lib_fd = pathmap_open(lib_name);
 971 
 972          if (lib_fd < 0) {
 973             print_debug("can't open shared object %s\n", lib_name);
 974             // continue with other libraries...
 975          } else {
 976             if (read_elf_header(lib_fd, &elf_ehdr)) {
 977                lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr);
 978                print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n",
 979                            lib_name, lib_base, lib_base_diff);
 980                // while adding library mappings we need to use "base difference".
 981                if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) {
 982                   print_debug("can't read shared object's segments\n");
 983                   close(lib_fd);
 984                   return false;
 985                }
 986                add_lib_info_fd(ph, lib_name, lib_fd, lib_base);
 987                // Map info is added for the library (lib_name) so
 988                // we need to re-sort it before calling the p_pdread.
 989                if (sort_map_array(ph) != true)
 990                   return false;
 991             } else {
 992                print_debug("can't read ELF header for shared object %s\n", lib_name);
 993                close(lib_fd);
 994                // continue with other libraries...
 995             }
 996          }
 997       }
 998 
 999     // read next link_map address
1000     if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET,
1001                    &link_map_addr, sizeof(uintptr_t)) != PS_OK) {
1002       print_debug("can't read next link in link_map\n");
1003       return false;
1004     }
1005   }
1006 
1007   return true;
1008 }
1009 
1010 // the one and only one exposed stuff from this file
1011 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
1012   ELF_EHDR core_ehdr;
1013   ELF_EHDR exec_ehdr;
1014   ELF_EHDR lib_ehdr;
1015 
1016   struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
1017   if (ph == NULL) {
1018     print_debug("can't allocate ps_prochandle\n");
1019     return NULL;
1020   }
1021 
1022   if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
1023     free(ph);
1024     print_debug("can't allocate ps_prochandle\n");
1025     return NULL;
1026   }
1027 
1028   // initialize ph
1029   ph->ops = &core_ops;
1030   ph->core->core_fd   = -1;
1031   ph->core->exec_fd   = -1;
1032   ph->core->interp_fd = -1;
1033 
1034   // open the core file
1035   if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
1036     print_debug("can't open core file\n");
1037     goto err;
1038   }
1039 
1040   // read core file ELF header
1041   if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) {
1042     print_debug("core file is not a valid ELF ET_CORE file\n");
1043     goto err;
1044   }
1045 
1046   if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
1047     print_debug("can't open executable file\n");
1048     goto err;
1049   }
1050 
1051   if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true ||
1052       ((exec_ehdr.e_type != ET_EXEC) && (exec_ehdr.e_type != ET_DYN))) {
1053     print_debug("executable file is not a valid ELF file\n");
1054     goto err;
1055   }
1056 
1057   // process core file segments
1058   if (read_core_segments(ph, &core_ehdr) != true) {
1059     goto err;
1060   }
1061 
1062   // process exec file segments
1063   if (read_exec_segments(ph, &exec_ehdr) != true) {
1064     goto err;
1065   }
1066 
1067   // exec file is also treated like a shared object for symbol search
1068   if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd,
1069                       (uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL) {
1070     goto err;
1071   }
1072 
1073   // allocate and sort maps into map_array, we need to do this
1074   // here because read_shared_lib_info needs to read from debuggee
1075   // address space
1076   if (sort_map_array(ph) != true) {
1077     goto err;
1078   }
1079 
1080   if (read_shared_lib_info(ph) != true) {
1081     goto err;
1082   }
1083 
1084   // sort again because we have added more mappings from shared objects
1085   if (sort_map_array(ph) != true) {
1086     goto err;
1087   }
1088 
1089   if (init_classsharing_workaround(ph) != true) {
1090     goto err;
1091   }
1092 
1093   return ph;
1094 
1095 err:
1096   Prelease(ph);
1097   return NULL;
1098 }